Onset of convection with temperature- and depth-dependent viscosity Jun Korenaga

advertisement
GEOPHYSICAL RESEARCH LETTERS, VOL. 29, NO. 19, 1923, doi:10.1029/2002GL015672, 2002
Onset of convection with temperature- and depth-dependent viscosity
Jun Korenaga
Department of Earth and Planetary Science, University of California, Berkeley, CA, USA
Thomas H. Jordan
Department of Earth Sciences, University of Southern California, Los Angeles, CA, USA
Received 13 June 2002; revised 29 July 2002; accepted 8 August 2002; published 9 October 2002.
[1] We derive a scaling law for the onset of convection in
an incompressible fluid cooled from above with
temperature- and depth-dependent viscosity, on the basis
of 2-D numerical simulation. Rayleigh numbers up to 107
are considered. The activation energy is varied from 0 to
200 kJ mol1, and a variation of up to 103 is considered in
depth-dependent reference viscosity. Our scaling law is
shown to be able to handle a variety of buoyancy
distributions, including highly non-Gaussian and
multimodal distributions, with a tightly constrained
INDEX TERMS: 3035 Marine
critical Rayleigh number.
Geology and Geophysics: Midocean ridge processes; 8121
Tectonophysics: Dynamics, convection currents and mantle
plumes; 8162 Tectonophysics: Evolution of the Earth:
Rheology—mantle. Citation: Korenaga, J., and T. H. Jordan,
Onset of convection with temperature- and depth-dependent
viscosity, Geophys. Res. Lett., 29(19), 1923, doi:10.1029/
2002GL015672, 2002.
1. Introduction
[2] The onset of convection in a fluid cooled from
above is a simple fluid dynamical problem, with profound
applications for the dynamics of the Earth’s mantle.
Because of the strong temperature dependency of mantle
rheology, most of temperature variations in mantle convection are confined in a cold and stiff boundary layer
(lithosphere), the dynamics of which dominates the global
energetics of mantle convection [e.g., Conrad and Hager,
1999]. The thickness of the boundary layer is one of
essential controlling parameters in such energetics. Thicker
lithosphere requires greater energy to be dissipated for its
deformation at subduction zones. The growth of the
boundary layer is limited by its intrinsic convective
instability, which is characterized by the onset of convection [e.g., Parsons and McKenzie, 1978; Yuen and Fleitout, 1984; Buck and Parmentier, 1986; Davaille and
Jaupart, 1994].
[3] A scaling law for the onset of convection has already
been established for the case of temperature-dependent
viscosity [Korenaga and Jordan, 2002]. The purpose of
this paper is to extend it to a more general case, i.e.,
temperature- and depth-dependent viscosity. This generalization is important because mantle melting beneath midocean ridges is expected to introduce intrinsic rheological
layering in the shallow upper mantle [e.g., Karato, 1986;
Hirth and Kohlstedt, 1996]. Residual mantle after melting,
Copyright 2002 by the American Geophysical Union.
0094-8276/02/2002GL015672$05.00
which normally occupies the upper 60 – 80 km of the
oceanic mantle, could be by two orders of magnitude more
viscous than unmelted mantle.
[4] Our approach is based on two-dimensional numerical simulation, in a manner similar to Korenaga and
Jordan [2002]. Depth-dependent viscosity introduces one
significant challenge for scaling analysis. Buoyancy density distribution becomes highly non-Gaussian and can
even become multimodal. We will demonstrate that the
concept of the differential Rayleigh number is still valid
for non-Gaussian buoyancy distribution, and also that
multimodal distribution can be treated successfully by
considering the coupling of sub-distributions.
2. Numerical Modeling
[5] The numerical formulation is basically the same as
that described by Korenaga and Jordan [2002], except for
the addition of depth-dependent viscosity (Figure 1). Only
key elements are summarized here. The governing equations for thermal convection of an incompressible fluid are
integrated by the finite element method. Length and time
are normalized by a system depth, D, and a diffusion time,
D2/k, respectively, where k is thermal diffusivity. Temperature is normalized by T, which is the difference
between the initial fluid temperature (T0) and the imposed
surface temperature (Ts). Viscosity is normalized by m0,
which is reference viscosity measured at the model base
with the initial temperature. The superscript * denotes
normalized variables. The aspect ratio of the model
domain is unity, with rigid horizontal and reflecting
vertical boundaries. The top and bottom temperature are
fixed at 0 and 1, respectively. The initial internal temperature is set to unity plus random perturbations with the
maximum amplitude of 105. The convection system has
one non-dimensional parameter called the Rayleigh number, which is defined as
Ra ¼
ar0 gTD3
km0
ð1Þ
where a is the coefficient of thermal expansion, g is
gravitational acceleration, and r0 is reference density at T0.
The following viscosity law is employed:
mðT *; z*Þ ¼ mr ð z*Þexp
E*
E*
;
* 1 þ Toff
*
T* þ Toff
ð2Þ
* = 273/T. E is activation
where E* = E/(RT) and Toff
energy, and R is the universal gas constant. T is set to
29 - 1
29 - 2
KORENAGA AND JORDAN: TEMPERATURE- AND DEPTH-DEPENDENT VISCOSITY
Figure 1. Model geometry and length scales of transient
cooling. Gray scale indicates temperature variation. Onset
of convection is significantly influenced by temperatureand depth-dependent viscosity.
1300. The z axis is downward. For the depth-dependent
part, we use a simple two-layer model as
mr ð z*Þ ¼
8
< mr
0 z* z*
b
:
z*
b < z* 1
1
ð3Þ
where mr is greater than unity. To ensure numerical
stability, the largest normalized viscosity is limited to 105.
[6] The range of activation energy used is 0– 30 kJ mol1
for Ra = 105, 0 – 120 kJ mol1 for Ra = 106, and 0 –200 kJ
mol1 for Ra = 107. An increment in activation energy is 10
for Ra = 105 and 106 and 20 for Ra = 107, which amounts to
28 different cases. Combined with all possible combinations
of z*b = 0.1, 0.25, and 0.5 and mr = 10, 102, and 103, we
conducted total 252 runs of the instantaneous cooling
model. The onset time of convection, t*c, is defined as when
a horizontally averaged temperature profile deviates from a
purely conductive profile by greater than 0.01 at any depth
level. Because of the rapid growth of convective instability,
our measurement is not very sensitive to this particular
value of the threshold. In the following, we focus on the
runs in which convection took place before a conductive
cooling front reaches the model bottom (i.e., t*c < 0.06), so
that the development of convective instability is least
influenced by finite-domain effects. The number of such
runs is 189.
[7] Our measurements of onset time are summarized in
Figure 2. Onset times, t*c, are normalized by the local
timescale for boundary-layer instabilities, t*r = Ra2/3, and
they are plotted as a function of E. The effect of depthdependent viscosity can be readily seen in the systematics of
onset time. Our task is to develop a general scaling law to
explain the observed systematics.
3. Scaling Analysis
[8] Our scaling analysis is based on the concept of the
differential Rayleigh number, which has been introduced by
Korenaga and Jordan [2002]. Because of intrinsic depthdependent viscosity, however, a similarity variable,
Figure 2. Summary of onset time measurements in terms of
the activation energy E and the square root of onset time (tc*)
scaled by local boundary-layer time scale (tr* = Ra2/3). Data
for purely temperature-dependent cases are from Korenaga
and Jordan [2002].
pffiffiffiffi
h ¼ z:*= 2 t* , cannot be used, and the present analysis
is formulated with two independent variables, i.e., z* and t*.
Following Howard [1966], we assume that convection takes
place when the local Rayleigh number reaches a critical
value as,
Rad ðtc*Þ ¼ Rc :
ð4Þ
To calculate Rad, we first define the distribution of available
buoyancy density as
bð z*; t*Þ ¼
1 dT *
:
m* dz*
ð5Þ
Some examples of b are shown in Figure 3. They are all
highly non-Gaussian, unlike the cases of temperature-
Figure 3. Examples of buoyancy density distribution
b(z*,t*c ). Case 1 (solid): Ra = 107, E = 20, z*b = 0.1, mr =
102, and t*c = 4.88 103. Case 2 (dashed): Ra = 106, E =
70, z*b = 0.25, mr = 103, and t*c = 3.29 102. Case 3
(dotted): Ra = 107, E = 80, z*b = 0.5, mr = 10, and t*c =
2.05 102.
KORENAGA AND JORDAN: TEMPERATURE- AND DEPTH-DEPENDENT VISCOSITY
29 - 3
dependent viscosity studied by Korenaga and Jordan
[2002]. The origin of available buoyancy may be calculated
as
zo* ¼ zm* b1 ðzm*; t*Þ
Z z*m
bð z*; t*Þdz*
ð6Þ
0
where b(zm*, t*) = max b(z*, t*). The differential Rayleigh
number is then defined as
dRa ¼
4ar0 gT ð z* zo*Þ3 D3 dT *
dz*:
km
dz*
ð7Þ
The local Rayleigh number is obtained by integrating this
differential Rayleigh number as
Z
1
Rad ðt*Þ ¼
dRa
ð8Þ
0
¼ Ra F ðt*Þ
ð9Þ
where F(t*) is a functional that depends on buoyancy
density as
Z
1
F ðt*Þ ¼
4ð z* zo*Þ3 bð z*; t*Þdz*:
ð10Þ
0
It can be shown that the functional F is approximately
(total available buoyancy) (width of buoyancy distribution)3. For a time-independent linear temperature profile
(i.e., dT*/dz* = 1) and uniform viscosity, F = 1. Our Rc is
related to the conventional critical Rayleigh number as Rac =
(p2/25)Rc [Korenaga and Jordan, 2002]. From equations (4)
and (9), therefore, we have
Rac ¼
p2
Ra F ðtc*Þ:
25
ð11Þ
The success of the scaling analysis may be measured by the
variance of Rac calculated from our modeling results using
equation (11).
[9] Estimated values for Rac, based on equations (6),
(10), and (11), are shown in Figure 4a. Most of estimates
fall in the range of 1500– 3000, which translates into about
±20% error in the onset time of convection. Some systematic trends can be seen, which are probably because we vary
the activation energy systematically, with the same random
ensemble for the initial temperature field. Several significant outliers are also observed, which correspond to the
cases with the bimodal distribution of buoyancy density.
These outliers clearly indicate that a multimodal distribution
must be treated with care. We may dissect a multimodal
distribution into non-overlapping N sub-distributions, each
with a single maximum, and calculate the functional F as
F ðt*Þ ¼
N Z zei*
X
i¼1
z*
si
4ð z* zoi*Þ3 bð z*; t*Þdz*;
ð12Þ
where z*si and z*ei define the range of the i-th subdistribution, and z*oi denotes its origin calculated in a
Figure 4. Estimated Rac as a function of onset time. (a)
Rac for unimodal buoyancy distributions (solid circle) are
confined in the rage of 1500– 3000. Outliers correspond
largely to bimodal distributions of buoyancy density. Open
and solid stars denote Rac calculated with equations (10)
and (12), respectively. (b) Rac becomes less scattered with
the consideration of buoyancy coupling [equation (14) with
p = 1.5]. Gray shadings show one and two standard
deviations.
similar manner to equation (6). This formula, however,
underestimates Rac, as shown in Figure 4a.
[10] This difficulty in defining a widely applicable functional F is rooted in the cubic dependence of the Rayleigh
number on the length scale. Simple summation of F values
corresponding to sub-distributions cannot represent the
system’s true convective instability. How, then, should we
proceed?
[11] One may realize that the above subdivision is a
rather arbitrary procedure. Consider, for example, a bimodal
distribution. If two maxima of the distribution are located
closely, with an insignificant minimum between them, the
distribution may reasonably be approximated as unimodal.
The functional F should not depend on whether this
distribution is treated as unimodal or bimodal. Equations
(10) and (12), however, give considerably different values.
On the other hand, if two maxima are located far apart with
a wide zero-buoyancy interval between them, we do not
29 - 4
KORENAGA AND JORDAN: TEMPERATURE- AND DEPTH-DEPENDENT VISCOSITY
expect that these two sub-distributions interact to produce
system-wide convective instability. The concept of subdistributions thus cannot be dropped. Adding up the convective instability of sub-distributions must be somehow
capable of bridging smoothly the above two extreme cases.
[12] The simplest remedy appears to consider some kind
of ‘coupling’ among sub-distributions. It is clear that such
coupling should be inversely proportional to the separation
between sub-distributions. An algebraic argument can also
show that the coupling may be implemented in terms of a
difference introduced by subdivision such as:
i ð z; t Þ ð z zo1 Þ3 ð z zoi Þ3
¼ 3ðzoi zo1 Þz 2 3 zoi2 z2o1 z þ z3oi z3o1 ;
ð13Þ
where zoi is a function of time. We thus define the functional
F for a multimodal distribution as
F ðt*Þ ¼
N Z zei*
X
i¼1
þ
z*
si
N
X
i¼2
4ð z* z*oiÞ3 bð z*; t*Þdz*
ð2s1 zsi* z*s1Þp
Z zei*
z*
si
i ð z*; t*Þbð z*; t*Þdz*;
ð14Þ
where s1 is one standard deviation of the first subdistribution, and p is an empirical exponent. We found
that, with p = 1.5, all of our onset-time measurements may
be explained by Ra = 2350 ± 330 (Figure 4b).
[13] Considering various types of buoyancy distributions
appeared in our modeling (e.g., Figure 3), the unifying
scaling law for the onset of convection [equations (11) and
(14)] can be considered to be quite successful. The wide
range of temperature- and depth-dependent viscosity
employed in this study ensures that our scaling law can
safely be applied to the convective instability of thermal
boundary layers in the Earth’s mantle.
[14] Acknowledgments. This work was sponsored in part by the
Miller Research Fellowship. We thank two anonymous reviewers for
helpful reviews.
References
Buck, W. R., and E. M. Parmentier, Convection beneath young oceanic
lithosphere: Implications for thermal structure and gravity, J. Geophys.
Res., 91, 1961 – 1974, 1986.
Conrad, C. P., and B. H. Hager, The thermal evolution of an earth with
strong subduction zones, Geophys. Res. Lett., 26, 3041 – 3044, 1999.
Davaille, A., and C. Jaupart, Onset of thermal convection in fluids with
temperature-dependent viscosity: Application to the oceanic mantle, J.
Geophys. Res., 99, 19,853 – 19,866, 1994.
Hirth, G., and D. L. Kohlstedt, Water in the oceanic mantle: Implications
for rheology, melt extraction, and the evolution of the lithosphere, Earth
Planet. Sci. Lett., 144, 93 – 108, 1996.
Howard, L. N., Convection at high Rayleigh number, in Proceedings of the
Eleventh International Congress of Applied Mechanics, edited by H.
Gortler. pp. 1109 – 1115, Springer-Verlag, New York, 1966.
Karato, S., Does partial melting reduce the strength of the upper mantle?,
Nature, 319, 309 – 310, 1986.
Korenaga, J., and T. H. Jordan, Physics of multi-scale convection in the
Earth’s mantle, 1, scaling laws, stability analysis, and 2-D numerical
results, in revision, J. Geophys. Res., 2002.
Parsons, B., and D. McKenzie, Mantle convection and the thermal structure
of the plates, J. Geophys. Res., 83, 4485 – 4496, 1978.
Yuen, D. A., and L. Fleitout, Stability of the oceanic lithosphere with
variable viscosity: an initial-value approach, Phys. Earth Planet. Int.,
34, 173 – 185, 1984.
J. Korenaga, Department of Earth and Planetary Science, University of
California, 377 McCone Hall, Berkeley, CA 94720-4767, USA. (korenaga@
seismo.berkeley.edu)
T. H. Jordan, Department of Earth Sciences, University of Southern
California, SCI 103, Los Angeles, CA 90089-0740, USA. (tjordan@usc.
edu)
Download